When Does Custom AI Become Counter-Productive?

So I notice I am routinely seeing “do you want a summary” features on news items I read. Many of those items are quite short, and I am starting to wonder whether the “summarize this” function adds any value at all when applied to short news items, as much as I might sometimes appreciate the feature in the context of reading lots of longer white papers. 

Which raises a logical question: Does every application, job function or process require its own dedicated AI? At what point does this become counter-productive? 

If a process is already efficient and user-friendly, then forcing AI integration might be counterproductive. AI arguably excels at tasks requiring complex data analysis, automation of repetitive actions, or handling large datasets.

Conversely, a task that is simple, unique and does not involve lots of data might not benefit much from AI, such as providing a summary of an already-short bit of text. 

Complexity creep is another possible downside. Think of a simple task requiring multiple AI handoffs, increasing troubleshooting complexity and potentially slowing down the process.

So less AI might be more. Having AI embedded in all software, for every function, for every job role might wind up being far less useful than is imagined. 

Adding such functionality also will tend to add cost. For simpler tasks, the cost might outweigh the benefit.

In one sense, much investment in artificial intelligence, perhaps best illustrated by generative AI, is going to be wasted. That is typical and normal for venture capital investments. 

But there might could be some developing parallels to the dotcom bubble of the late 1990s, when too much money chased too few good ideas, resulting in over-investment, as VC Josh Elman talks about it. 

 

source: Wall Street Journal

Past is not inevitably prologue, so there can be reasonable hope that only typical VC failure rates will happen with generative AI, and not the excesses of the dotcom bubble, when unrealistic expectations led to over-investment in unsustainable business models and inflated valuations of companies with little to no revenue or clear paths to profitability.

In many cases, there was a  "Build It and They Will Come" mentality that downplayed creation of compelling value propositions and establishing a sustainable business model. 

And, to be sure, many ask questions about generative AI revenue models. But that is a logical kind of question that was not always asked when startups were building their firms. 

One reason it often is difficult to grasp how artificial intelligence will appear is that it can be manifest in so many different ways. It can take the form of a discrete app; a capability; a platform or, eventually, possibly something else. 

AI can function as a specific skill or ability embedded within a larger system, such as facial recognition software within images or videos. In this case, AI isn't a standalone entity but acts as a crucial component of the software's functionality.

AI also can take the form of dedicated applications designed to perform specific tasks. Examples include chatbots trained to answer customer service inquiries, language translation apps that leverage AI for real-time communication, or even mobile navigation systems that employ AI for route optimization. Many vertical apps designed for specific industries will provide examples. 

AI can evolve into platforms that provide the foundation for developing and deploying various AI-powered applications. TensorFlow, PyTorch, and Microsoft Azure Cognitive Services are prominent examples. These platforms offer tools, resources, and infrastructure that software developers can utilize to create and integrate AI functionalities within their applications.

It might also become “something else” that blurs the lines between tool, companion or “extension of self.”